One paper is accepted to IEEE Access

With Junhee Cho and Seonghak Lee; Prof. Jongmin Lee and Prof. Junseok Kwon served as co-corresponding authors. DiNGGA reconstructs fine-grained facial detail in Gaussian avatars by combining Dirichlet-based part guidance with noise-guided optimization, improving the fidelity of subtle facial regions that standard 3DGS avatars tend to blur.